Scientists endeavor to discover new semiconductor materials that could stimulate the effectiveness of solar cells and other electronics. But the pace of innovation is a bottleneck by speed at which researchers can manually measure the properties of important materials.
A fully autonomous robotic system developed by MIT researchers could accelerate things.
Their system uses a robotic probe to measure an important electrical property known as photoconductance, which is to what extent an electrical material is reactive to the presence of light.
Researchers inject the knowledge of the Domaine materials of human experts in the automatic learning model that guides the robot decision-making. This allows the robot to identify the best places to contact equipment with the probe to obtain the most information on its photoconductor, while a specialized planning procedure finds the fastest way to move between the contact points.
During a 24 -hour test, the completely autonomous robotic probe has taken more than 125 unique measures per hour, with more precision and reliability than other methods based on artificial intelligence.
By considerably increasing the speed at which scientists can characterize the important properties of new semiconductor materials, this method could stimulate the development of solar panels that produce more electricity.
“I find this Paper to be incredibly exciting because it provides a pathway for autonomous, contact-based characterization methods. Not every important owner of a material can be measured in a contactless way. If you need to make contact with your sample, you want it to be to be Maximize the Amount of Information that you gain, ”Says Tonio Buonassisi, Professor of Mechanical Engineering and Senior Author of A paper on the autonomous system.
His co-authors include the main author Alexander (Aleks) Siemenn, a graduate student; Postdocs Basita Das and Kangyu Ji; and Fang Sheng graduate student. The work appears today in Scientific advances.
Contact
Since 2018, researchers from the Buonassisi laboratory have been working towards a completely autonomous material discovery laboratory. They recently focused on the discovery of new perovskites, which are a class of semiconductor materials used in photovoltaics such as solar panels.
In previous work, they have developed techniques to synthesize and quickly print unique combinations of perovskite materials. They also designed Methods based on imaging To determine certain important material properties.
But photoconductor is most precisely characterized by placing a probe on the material, brilliantly a light and measuring the electrical response.
“To allow our experimental laboratory to operate as quickly and precisely as possible, we had to find a solution that would produce the best measures while minimizing the time necessary to perform the entire procedure,” explains Siemennn.
This required the integration of automatic learning, robotics and materials science in an autonomous system.
To start, the robotic system uses its on -board camera to take an image of a slide with perovskite material printed on it.
Then, he uses computer vision to cut this image in segments, which are introduced into a neural network model specially designed to incorporate the expertise in the field of chemists and materials of materials.
“These robots can improve the repeatability and precision of our operations, but it is important to have a human in the loop. If we do not have a good way to implement the rich knowledge of these chemical experts in our robots, we will not be able to discover new materials, ”adds Siemenn.
The model uses this knowledge of the domain to determine the optimal points for the probe to contact according to the shape of the sample and its material composition. These contact points are introduced into a path planner who finds the most effective way for the probe to reach all points.
The adaptability of this automatic learning approach is particularly important because printed samples have unique forms, circular drops with Jellybean type structures.
“It's almost like measuring snowflake – it is difficult to get two that are identical,” says Buonassisi.
Once the path planner finds the shortest path, he sends signals to the robot engines, which handle the probe and take measures at each rapid succession point.
The key to the speed of this approach is the self-supervised nature of the neural network model. The model determines the optimal contact points directly on an image example – without the need for labeled training data.
Researchers have also accelerated the system by improving the path planning procedure. They found that adding a small amount of noise or chance to the algorithm helped him find the shortest way.
“As we progress in this era of autonomous laboratories, you really need these three expertise – creation of hardware, software and understanding of materials science – meeting in the same team to be able to innovate quickly. And that is part of the secret sauce here, “says Buonassisi.
Rich data, fast results
Once they built the system from zero, the researchers tested each component. Their results have shown that the neural network model found better contact points with less calculation time than seven other AI -based methods. In addition, the path planning algorithm has constantly found shorter plans than other methods.
When they gathered all the parts to carry out an entirely autonomous 24 -hour experience, the robotic system has carried out more than 3,000 unique photoconductor measurements at a rate greater than 125 an hour.
In addition, the level of detail provided by this precise measurement approach allowed researchers to identify the hot spots with higher photoconduction as well as fields of degradation of materials.
“The possibility of collecting such rich data which can be captured at such fast rhythms, without needing human advice, begins to open doors to be able to discover and develop new high performance semiconductors, in particular for sustainability applications such as solar panels,” explains Siemenn.
The researchers want to continue to rely on this robotic system as they strive to create a completely autonomous laboratory for the discovery of materials.
This work is supported, in part, by First Solar, eni by MIT Energy Initiative, Mathworks, the Acceleration Consortium of the University of Toronto, the American Department of Energy and the National American Science.
